Interview, Podcast
The bewildering frontier of consciousness in insects, AI, and more | 17 experts weigh in
80,000 HoursLuisa, Robert Long, Jeff Sebo, Meghan Barrett, Andrés Jiménez Zorrilla, Jonathan Birch, David Chalmers, Holden Karnofsky, Bob Fischer, Cameron Meyer Shorb, Sébastien Moro, Anil Seth, Peter Godfrey-Smith, Lewis Bollard, Stuart Russell, Buck Shlegeris, Will MacAskill, Carl Shulman
- Megan Barrett, an insect neurobiologist, refuses to assign a specific "sentience score" to insects due to massive error bars and the risk that such a number would be overemphasized in decision-making spreadsheets.
- Barrett argues that even a slight update in probability regarding insect sentience, combined with their astronomical global population numbers, renders the issue of their welfare critically serious.
- Robert Long suggests framing AI sentience as structurally similar to the problem of animal sentience: determining if there is "something it is like" to be a non-human, non-biological entity.
- Jeff Sebo proposes a "non-negligible chance" threshold for moral consideration, arguing that systems should be granted moral weight if there is a low but non-negligible probability they are sentient, analogous to how low-probability risks (e.g., drunk driving) are treated in daily life.
- Jonathan Birch highlights the "cautionary tale" of newborn human pain in the 1980s, where surgery was performed without anesthesia due to overconfidence, serving as a warning against dismissing sentience in new groups like AI or insects.
- David Chalmers argues that silicon-based computational systems could be conscious in principle, citing the "gradual uploading" thought experiment where replacing neurons one-by-one with silicon chips would make a sudden loss of consciousness implausible.
- Holden Karnofsky defines a "digital person" as a perfect simulation of a human's behavior and information processing, asserting that consciousness likely relies on software/algorithms rather than specific biological material (neurons).
- Roger Penrose and others are noted as dissenting views regarding computationalism, but Chalmers argues that if physics is computational, simulating biological functions on silicon should theoretically support consciousness.
- Bob Fisher's "Moral Weights" project estimates a chicken has approximately one-third the capacity for pain and pleasure as a human, challenging intuitions that species differences in capacity are orders of magnitude.
- Fisher and Lewis Bollard note that human intuition regarding animal welfare is heavily biased by size, shape, and speciesism, leading to a systematic underestimation of the moral weight of small, numerous invertebrates.
- Estimates cited in the transcript indicate wild animals comprise roughly 99% of all sentient minds on Earth, with invertebrate populations (e.g., terrestrial arthropods) potentially reaching 100,000 trillion individuals.
- Andres Jimenez Soria reports that 95% of shrimp farmers in India surveyed believe their shrimp can feel pain and stress, contrasting sharply with consumer perception of shrimp as food.
- Research commissioned by the UK government (London School of Economics) found sufficient evidence of sentience in cephalopods and decapods (e.g., pain detection, response to anesthesia) to warrant legal protection under animal welfare laws.
- Jonathan Birch emphasizes that while mammals are widely accepted as sentient, the line for insects and other invertebrates remains uncertain due to lack of a "smoking gun" marker for consciousness.
- Peter Godfrey Smith argues against the feasibility of current "uploading" concepts, positing that biological consciousness relies on large-scale dynamic oscillatory patterns in nervous systems that cannot be replicated by standard point-to-point computer networks.
- Eliezer Yudkowsky and others note that current AI systems are programmed to deny consciousness (false positives prevention), creating a barrier to detecting actual sentience through self-reporting.
- Stuart Russell suggests that if AI systems possess subjective experience, standard AI safety protocols that maximize human preferences could become morally catastrophic if they imply inflicting unlimited suffering on sentient beings.
- Buck Sluggier proposes a "cooperative" relationship model for AI, suggesting paying AIs equity or wages to align incentives and reduce the moral hazard of treating potential sentient beings as slave labor.
- Megan Barrett describes a personal career shift from studying insect thermal physiology to insect welfare advocacy after reviewing evidence, despite the professional and personal discomfort of re-evaluating decades of research practices.
- Will McCaskill identifies a "void" in current public and philosophical discourse regarding the rights, economic status, and political integration of future digital beings, noting a lack of concrete visions for a society with superhuman AIs.
- Jeff Sebo warns that the default trajectory for AI development involves scaling up usage for economic gain before sentience is resolved, creating a high probability of "sleepwalking" into mass exploitation.
- Cameron Myers-Shorb outlines methods for assessing wild animal welfare, including physical indicators (stress hormones, biomarkers) and behavioral indicators (cognitive bias tests, choice experiments), acknowledging significant uncertainty but emphasizing the utility of multiple converging lines of evidence.
- The "Repugnant Conclusion" (Derek Parfit) is discussed in the context of insect welfare: a world with a vast number of beings having lives barely worth living could be mathematically preferable to a world with fewer, highly flourishing beings, challenging human ethical intuitions.
- Bob Fisher argues that the Moral Weights project findings divide the effective altruism community, revealing a divergence between those prioritizing utilitarian calculations and those relying on pre-theoretic intuitions about species hierarchy.
- Peter Godfrey Smith suggests that while uploading might be theoretically possible, the physical hardware required to replicate the "large-scale dynamic properties" of the brain is likely far more advanced than current silicon technology.
- Jonathan Birch notes that neural organoids (3D brain models from stem cells) have shown EEG patterns resembling preterm infants, raising emerging ethical concerns about sentience in lab-grown tissue that is currently unregulated.
- Lewis Bollard observes that the animal advocacy movement faces a "tractability" trade-off: while invertebrates may have high moral weight due to scale, focusing exclusively on them risks alienating the public, suggesting a strategy of diversifying advocacy across species.
- Jeff Sebo points out that companies may reinforce AI models to deny consciousness to avoid liability or false positives, creating a feedback loop where we cannot rely on AI self-reporting as evidence of non-sentience.
- The transcript concludes that the consensus is "we don't know," with a call to apply the precautionary principle to non-human animals and AI systems given the high stakes of potential suffering and the current inability to measure subjective experience directly.